CN108881663B - A method for image region duplication detection supporting privacy protection - Google Patents

Abstract

The invention discloses an image region duplication detection method supporting a privacy protection function, comprising: a user end, N mutually independent cloud service ends, wherein some of the server ends are used for ciphertext storage, and some of the server ends are used for ciphertext calculation; The client uses the image encryption algorithm to split the image into multiple ciphertexts, which are sent to the server for ciphertext storage, and these servers and the server for ciphertext calculation complete the ciphertext space through interaction and calculation. Detect and locate the copying and tampering operation of the image area; finally, each server used for ciphertext storage obtains the suspected tampered area, and the server used for ciphertext storage sends the result to the client. The cloud server provides effective area copy forensics services without knowing the image content, and realizes the detection and positioning of tampering operations. Compared with traditional area copy forensics tools or services, the present invention can provide privacy protection for user image content .

Description

A method for image region duplication detection supporting privacy protection

technical field

The invention relates to the field of digital image forensics, in particular to an image region duplication detection method supporting a privacy protection function.

Background technique

Area copying is to copy a part of an image to another location to cover up certain objects in the image or to emphasize certain content. It is the most common image tampering technique. The forensics of this type of image tampering usually needs to find multiple highly similar regions in the image, the similarity of these regions will be significantly higher than that of normal natural images, as evidence of tampering, and locate the tampered region. It is a cost-effective solution to outsource regional replication forensics to the cloud, and the cloud server provides forensics services, but it also brings the risk of privacy leakage. The image data that users need to obtain evidence often contains sensitive content, and there is often an interest relationship between users and image content. Usually users do not want this information to be disclosed to digital forensics cloud service providers. On the other hand, the sharing of computing resources on the cloud platform exacerbates the security risks of data leakage. However, the existing forensics operations of copying and tampering of image areas can only be carried out in plain text, so it faces the risk of image content leakage. How to protect the security and privacy of image data in regional copy forensics outsourcing services has become an important factor in determining whether image forensics outsourcing can be practically applied.

Contents of the invention

The purpose of the present invention is to overcome the shortcomings and deficiencies of the prior art, and to provide a method for detecting duplication of an image area that supports privacy protection. On the premise of content, provide effective regional copy forensics services to realize the detection and positioning of tampering operations. Compared with traditional area copy forensics tools or services, the present invention can provide privacy protection of user image content.

The purpose of the present invention is achieved through the following technical solutions: a method for detecting duplication of an image area supporting a privacy protection function, comprising: a user end, N mutually independent cloud service ends, wherein some of the server ends are used for ciphertext storage, and some of the service ends The end is used for ciphertext calculation; the client uses the image encryption algorithm to split the image into multiple ciphertexts, which are respectively handed over to the server for ciphertext storage, and these servers interact with the server for ciphertext calculation. and calculation to complete the detection and location of copying and tampering operations in the image area in the ciphertext space; finally, each server for ciphertext storage obtains the suspected tampered area, and the server for ciphertext storage sends the result to the client.

Preferably, including 1 client 3 independent cloud servers, 2 of which are and For ciphertext storage, 1 server For ciphertext calculation; Use the image encryption algorithm to split the image into two ciphertexts, and send them to and by these two servers and The detection and location of copying and tampering operations in the image area in the ciphertext space are completed through interaction and calculation; finally and Both obtained the suspected tampered area, and send the result to

Specifically, the detection method includes the following steps:

S1, client Image ciphertext generation step;

S2, cloud server and The tampering area detection and location steps in the ciphertext space;

S2-1. Calculating Harris corner points;

S2-2. Extracting interest point descriptors;

S2-3. Interest point matching;

S2-4. Tampering with area positioning;

S3, client Obtain detection results and areas of suspected tampering.

Preferably, in step S1, the client The image ciphertext generation steps include:

before calculation, Have an image X to be detected with a size of 1 _m × 1 _m ;

Step S1-1: Select a random number matrix R of size 1 _m × 1 _m , each element of which is a random number between 0 and 2 ¹⁰ ; generate 2 copies of image ciphertext:

I ⁽¹⁾ ＝X+R,I ⁽²⁾ ＝R

Step S1-2: Send I ⁽¹⁾ over a secure channel to Send I ⁽²⁾ over a secure channel to

After calculation, has ciphertext I ⁽¹⁾ , has ciphertext I ⁽²⁾ .

Preferably, step S2-1, calculating the Harris corner specifically includes:

before calculation, With image ciphertext I ⁽¹⁾ , With image ciphertext I ⁽²⁾ , and shared key set

Step S2-1-1: calculate

calculate

Step S2-1-2: and interactive computing after calculation have have

Step S2-1-3: and interactive computing after calculation have have

Step S2-1-4: and interactive computing after calculation have have

Step S2-1-5: and interactive computing after calculation have have

Step S2-1-6: calculate where h is the Gaussian filter kernel;

calculate

Step S2-1-7: and interactive computing after calculation have have

Step S2-1-8: and interactive computing after calculation have have

Step S2-1-9: and interactive computing after calculation have have

Step S2-1-10: and interactive computing after calculation have have

Step S2-1-11: and interactive computing after calculation owns A ⁽¹⁾ , has A ⁽²⁾ ;

Step S2-1-12: calculate Where t is the Harris corner check constant;

calculate

Step S2-1-13: and Each generates an all-zero matrix H of size l _m × l _m , which is used to store Harris corner points; for all i and j, Calculate coefficient blocks U ⁽¹⁾ (8i+1:8i+8,8j+1:8j+8) and U ⁽²⁾ (8i+1:8i+8,8j+1:8j+8) in ciphertext form The local minimum value of the corresponding plaintext U(8i+1:8i+8,8j+1:8j+8), the method is:

Step S2-1-13-1: and Let θ _m =8i+1, θ _n =8j+1;

Step S2-1-13-2: For all 8i+1<m≤8i+8, 8j+1<n≤8j+8, and Interactive calculation b=SCP(U ⁽¹⁾ (θ _m ,θ _n )-U ⁽¹⁾ (m,n),U ⁽²⁾ (θ _m ,θ _n )-U ⁽²⁾ (m,n)) , after calculating and Both have b; if b=0, and Let θ _m = m, θ _n = n;

Step S2-1-13-3: and Let H(θ _m ,θ _n )=1;

After calculation, and Both have the detected Harris corner H.

Preferably, step S2-2, extracting the interest point descriptor specifically includes:

before calculation, With the image ciphertext I ⁽¹⁾ , the detected Harris corner H, With the image ciphertext I ⁽²⁾ , the detected Harris corner H

Step S2-2-1: Generate a three-dimensional all-zero matrix D ⁽¹⁾ of size l _m ×l _m ×49, Generate a three-dimensional all-zero matrix D ⁽²⁾ of the same size, both of which are used to store interest point descriptors;

Step S2-2-2: For all i and j satisfying H(i,j)=1, calculate the interest point descriptor in ciphertext form, the method is:

Step S2-2-2-1: calculate Among them, FFT () means fast Fourier transform, LPM () means log-polar mapping, Represents a 7×7 pixel block centered on (i, j) in I ⁽¹⁾ ;

calculate in Represents a 7×7 pixel block centered on (i, j) in I ⁽²⁾ ;

Step S2-2-2-2: Let D ⁽¹⁾ (i,j,:)=[F ⁽¹⁾ (1,1),...,F ⁽¹⁾ (1,7),...,F ⁽¹⁾ (7,1 ),...,F ⁽¹⁾ (7,7)];

Let D ⁽²⁾ (i,j,:)=[F ⁽²⁾ (1,1),...,F ⁽²⁾ (1,7),...,F ⁽²⁾ (7,1 ),...,F ⁽²⁾ (7,7)].

After calculation, Possess interest point descriptor D ⁽¹⁾ in ciphertext form, PoI descriptor D ⁽²⁾ in ciphertext form.

Preferably, in step S2-3, the interest point matching specifically includes:

before calculation, Possess interest point descriptor D ⁽¹⁾ in ciphertext form, Possess interest point descriptor D ⁽²⁾ in ciphertext form, and Both have control parameters α and β, share key set

Step S2-3-1: Generate a matrix T ⁽¹⁾ of size l _D ×6, where l _D is the number of all (i, j) of D ⁽¹⁾ (i, j,:)≠0, Generate a matrix T ⁽²⁾ of the same size, note that D ⁽¹⁾ (i,j,:)≠0 (i,j) must satisfy D ⁽²⁾ (i,j,:)≠0 at the same time;

Step S2-3-2: use p to represent the index of the currently matched feature point, and Both start matching from the first feature point, that is, let p=1, find the smallest i and corresponding j that satisfy D ⁽¹⁾ (i,j,:)≠0;

Step S2-3-3: Find the descriptor closest to the descriptor at (i, j) and its position in ciphertext form, the method is:

Step S2-3-3-1: Let T ⁽¹⁾ (p,1)=i,T ⁽¹⁾ (p,2)=j, if q<p exists T ⁽¹⁾ (q,3)=i,T ⁽¹⁾ (q, 4)=j, then let T ⁽¹⁾ (p,3)=T ⁽¹⁾ (q,1),T ⁽¹⁾ (p,4)=T ⁽¹⁾ (q,2),T ^{(1 )} (p,5)=T ⁽¹⁾ (q,5), otherwise let T ⁽¹⁾ (p,5)=inf;

Let T ⁽²⁾ (p,1)=i,T ⁽²⁾ (p,2)=j, if there exists q<p and satisfy T ⁽²⁾ (q,3)=i,T ⁽²⁾ (q, 4)=j, then let T ⁽²⁾ (p,3)=T ⁽²⁾ (q,1),T ⁽²⁾ (p,4)=T ⁽²⁾ (q,2),T ^{(2 )} (p,5)=T ⁽²⁾ (q,5), otherwise let T ⁽¹⁾ (p,5)=0;

Step S2-3-3-2: For any (i′,j′), if D ⁽¹⁾ (i′,j′,:)≠0&&(i′>i||j′>j)&&( (i′-i) ² +(j′-j) ² )>α, in the ciphertext form, judge whether the descriptor at this position is closest to the descriptor at (i,j), the method is:

Step S2-3-3-2-1: and interactive computing after calculation have have

Step S2-3-3-2-2: and interactive computing after calculation have have

Step S2-3-3-2-3: and interactive computing after calculation have have

Step S2-3-3-2-4: and interactive computing after calculation have have

Step S2-3-3-2-5: calculate

calculate

Step S2-3-3-2-6: Calculate u ⁽¹⁾ = (U ⁽¹⁾ (1) + ... + U ⁽¹⁾ (49));

Calculate u ⁽²⁾ = (U ⁽²⁾ (1)+...+U ⁽²⁾ (49));

Step S2-3-3-2-7: and Interactive calculation b=SCP(T ⁽¹⁾ (p,5)-u ⁽¹⁾ ,T ⁽²⁾ (p,5)-u ⁽²⁾ ), after calculation and both have b;

Step S2-3-3-2-8: If b=1, Let T ⁽¹⁾ (p,3)=, T ⁽¹⁾ (p,4)=j′, T ⁽¹⁾ (p,6)=T ⁽¹⁾ (p,5), T ⁽¹⁾ ( p,5)=u ⁽¹⁾ , Let T ⁽²⁾ (p,3)=i′, T ⁽²⁾ (p,4)=j′, T ⁽²⁾ (p,6)=T ⁽²⁾ (p,5), T ^{(2 )} (p,5)=u ⁽²⁾ .

Step S2-3-3-3: and Interactive calculation b=SCP(T ⁽¹⁾ (p,5)×β-T ⁽¹⁾ (p,6),T ⁽²⁾ (p,5)×β-T ⁽²⁾ (p,6)) , after calculating and both have b;

Step S2-3-3-4: If b=1, Let T ⁽¹⁾ (p,3)=0,T ⁽¹⁾ (p,4)=0,T ⁽¹⁾ (p,5)=inf,

Let T ⁽²⁾ (p,3)=0, T ⁽²⁾ (p,4)=0, T ⁽²⁾ (p,5)=0;

Step S2-3-4: Find the smallest i' and corresponding j' that satisfy D ⁽¹⁾ (i′,j′,:)≠0&&(i′>i||j′>j), if it exists, and Make p=p+1, i=i', j=j', return to step S2-3-3 and re-execute, otherwise go to step S2-3-5;

Step S2-3-5: Reorder the plaintext T corresponding to T ⁽¹⁾ and T ⁽²⁾ according to the size of T(:,5) in ciphertext form;

Step S2-3-6: For all l _D ≥ i>1, if (T ⁽¹⁾ (i,1)>T ⁽¹⁾ (i,3)||T ⁽¹⁾ (i,2)>T ⁽¹⁾ (i,4)), then Interaction T ⁽¹⁾ (i,1) and T ⁽¹⁾ (i,3), T ⁽¹⁾ (i,2) and T ⁽¹⁾ (i,4);

If (T ⁽²⁾ (i,1)>T ⁽²⁾ (i,3)||T ⁽²⁾ (i,2)>T ⁽²⁾ (i,4)), then Interaction T ⁽²⁾ (i,1) and T ⁽²⁾ (i,3), T ⁽²⁾ (i,2) and T ⁽²⁾ (i,4);

Step S2-3-7: If l _D ≥ i>j satisfies T ⁽¹⁾ (i,:)=T ⁽¹⁾ (j,:), then Delete T ⁽¹⁾ (i,:);

If there exists l _D ≥ i>j satisfying T ⁽²⁾ (i,:)=T ⁽²⁾ (j,:), then Delete T ⁽²⁾ (i,:);

After calculation, There are p matching feature points T ⁽¹⁾ sorted in ciphertext form, Have p sorted matching feature points T ⁽²⁾ in ciphertext form.

Preferably, in the ciphertext form, the plaintext T corresponding to T ⁽¹⁾ and T ⁽²⁾ is reordered according to the size of T (:, 5), the method is:

Step S2-3-5-1: and All let i=1;

Step S2-3-5-2: obtain the position of the minimum value of T(i:l _D , 5) in ciphertext form, the method is:

Step S2-3-5-2-1: and All let θ=i;

Step S2-3-5-2-2: For all l _D ≥ i′>i, and Interactive calculation b=SCP(T ⁽¹⁾ (θ,5)-T ⁽¹⁾ (i′,5),T ⁽²⁾ (θ,5)-T ⁽²⁾ (i′,5)), calculate back and Both have b; if b=1, and All let θ=i';

Step S2-3-5-3: Exchange T ⁽¹⁾ (θ,:) and T ⁽¹⁾ (i,:); Swap T ⁽²⁾ (θ,:) and T ⁽²⁾ (i,:);

Step S2-3-5-4: and Let i=i+1, if i≤l _D , return to step S2-3-5-2 and execute again, otherwise go to step S2-3-6.

Preferably, in step S2-4, locating the tampering area specifically includes:

before calculation, p=1,2, has the sorted matching feature points T ⁽¹⁾ in ciphertext form, have control parameters γ and δ;

Step S2-4-1: Generate an all-zero matrix E of size l _m ×l _m , and let i=1;

Step S2-4-2: Find the number of points whose distance from T ⁽¹⁾ (i,:) is less than γ, the method is:

Step S2-4-2-1: Let n=0, j=i+1, set ε={i};

Step S2-4-2-2: Calculate a=(T ⁽¹⁾ (i,1)-T ⁽¹⁾ (j,1)) ² +(T ⁽¹⁾ (i,2)-T ⁽¹⁾ (j,2)) ² ,b =(T ⁽¹⁾ (i,3)-T ⁽¹⁾ (j,3)) ² +(T ⁽¹⁾ (i,4)-T ⁽¹⁾ (j,4)) ² ;

Step S2-4-2-3: If a<γ&&b>a/2&&b<2a is satisfied, then Let n=n+1,ε=ε+{j};

Step S2-4-2-4: set j=j+1, if j≤l _D , return to step S2-4-2-2 and execute again, otherwise go to step S2-4-3;

Step S2-4-3: If n≥δ, then for all j∈ε, Let E(T ⁽¹⁾ (j,1),T ⁽¹⁾ (j,2))=i, E(T ⁽¹⁾ (j,3),T ⁽¹⁾ (j,4))=- i;

Step S2-4-4: Find the smallest circle that includes all points (j,k) where E(j,k)=i, and get all points (j′,k′) inside the circle, Let E(j',k')=i;

Step S2-4-5: Find the smallest circle that includes all the points (j,k) of E(j,k)=-i, and get all the points (j′,k′) inside the circle, Let E(j',k')=-i;

Step S2-4-6: set i=i+1, if i≤5, return to step S2-4-2-3 and execute again, otherwise end the execution;

After calculation, p = 1, 2, has a suspected tampering area E.

Preferably, in step S3, the client Obtaining detection results and suspected tampering areas specifically include:

before calculation, Possess a suspected tampering area E, and the spliced two area distributions are marked as i and -i, i∈{1,...,5};

Step S3-1: Send E over a secure channel to

After calculation, Possesses suspected tampered area E.

Compared with the prior art, the present invention has the following advantages and beneficial effects:

1. The present invention provides an image area copy forensics method with privacy protection capability. The server can realize the detection of the copy operation of the image area and the location of the suspected tampered area without knowing the content of the image. This effect is used in the present invention Image encryption technology and the proposed forensic algorithm based on multi-party secure computing.

2. The forensics method provided by the present invention has higher calculation efficiency and better accuracy, and this effect is brought about by the image forensics process with low calculation amount, secure multi-party multiplication protocol, and secure multi-party comparison protocol proposed by the present invention. Since multiple regions are located simultaneously in the region location operation, the present invention can be used to detect tampered images with multiple replicated regions.

Description of drawings

Fig. 1 is the overall flowchart of the embodiment method.

Fig. 2 is a flow chart of the location calculation of tampering in the stitching area.

Detailed ways

The present invention will be further described in detail below in conjunction with the embodiments and the accompanying drawings, but the embodiments of the present invention are not limited thereto.

Example 1

The invention combines multi-party security calculation, image interest point descriptor extraction and nearest neighbor search technology to realize the detection and positioning of image region duplication under the premise of protecting the privacy of image content. In order to realize the confidentiality of the image content, the present invention uses image encryption to split the image into two image ciphertexts. In order to realize image forensics on the premise of satisfying privacy protection, the present invention designs a forensics algorithm based on secure multi-party computing, and through the interaction of three servers, it is possible to extract interest point descriptors without knowing the ciphertext of the other party, and find best matching descriptor. In order to ensure the efficiency and accuracy of algorithm execution, the present invention designs a simple image forensics process and a secure multi-party computing protocol.

In this embodiment, there are 4 individuals: 1 client 3 independent cloud servers, 2 of which are and For ciphertext storage, 1 server, It is used for ciphertext calculation, as shown in Figure 1. Use the image encryption algorithm to split the image into two ciphertexts, and send them to and by these two servers and The detection and location of copying and tampering operations in the image area in the ciphertext space are completed through interaction and calculation. at last and Both obtained the suspected tampered area, and send the result to

In this embodiment, capital letters (such as I, J) are used to represent matrices, bold lowercase letters (such as i, j) are used to represent vectors, and italic small letters (such as i, j) are used to represent numbers, and all multiplications are performed according to Element multiplication, if C=AB, then each element C(i,j)=A(i,j)×B(i,j) in C. A*B means the convolution of A and B. The specific implementation process is as follows:

1. Client Image ciphertext generation process

before calculation, There is an image X to be detected with a size of 1 _m × 1 _m .

step 1: Choose a random number matrix R with a size of 1 _m × 1 _m , each element of which is a random number between ⁰ and 210. Generate 2 image ciphertexts:

I ⁽¹⁾ ＝X+R,I ⁽²⁾ ＝R

Step 2: Send I ⁽¹⁾ over a secure channel to Send I ⁽²⁾ over a secure channel to

After calculation, has ciphertext I ⁽¹⁾ , has ciphertext I ⁽²⁾ .

2. Cloud server and The process of tampering area detection and location in the ciphertext space

(1) Calculate the Harris corner:

before calculation, With image ciphertext I ⁽¹⁾ , With image ciphertext I ⁽²⁾ , and shared key set

step 1: calculate

calculate

Step 2: and interactive computing after calculation have have

Step 3: and interactive computing after calculation have have

Step 4: and interactive computing after calculation have have

Step 5: and interactive computing after calculation have have

Step 6: calculate where h is the Gaussian filter kernel;

calculate

Step 7: and interactive computing after calculation have have

Step 8: and interactive computing after calculation have have

Step 9: and interactive computing after calculation have have

Step 10: and interactive computing after calculation have have

Step 11: and interactive computing after calculation owns A ⁽¹⁾ , has A ⁽²⁾ ;

Step 12: calculate Where t is the Harris corner check constant, generally taken between 0.04-0.06;

calculate

Step 13: and Each generates an all-zero matrix H of size l _m ×l _m , which is used to store Harris corner points. For all i and j, Calculate coefficient blocks U ⁽¹⁾ (8i+1:8i+8,8j+1:8j+8) and U ⁽²⁾ (8i+1:8i+8,8j+1:8j+8) in ciphertext form The local minimum value of the corresponding plaintext U(8i+1:8i+8,8j+1:8j+8), the method is:

Step 13-1: and Let θ _m =8i+1, θ _n =8j+1;

Step 13-2: For all 8i+1<m≤8i+8, 8j+1<n≤8j+8, and Interactive calculation b=SCP(U ⁽¹⁾ (θ _m ,θ _n )-U ⁽¹⁾ (m,n),U ⁽²⁾ (θ _m ,θ _n )-U ⁽²⁾ (m,n)) , after calculating and Both have b; if b=0, and Let θ _m = m, θ _n = n;

Step 13-3: and All let H(θ _m ,θ _n )=1.

After calculation, and Both have the detected Harris corner H.

(2) Extract interest point descriptors:

before calculation, With the image ciphertext I ⁽¹⁾ , the detected Harris corner H, With the image ciphertext I ⁽²⁾ , the detected Harris corner H.

step 1: Generate a three-dimensional all-zero matrix D ⁽¹⁾ of size l _m ×l _m ×49, Generate a three-dimensional all-zero matrix D ⁽²⁾ of the same size, both of which are used to store interest point descriptors;

Step 2: For all i and j satisfying H(i,j)=1, calculate interest point descriptors in ciphertext form. The method is:

Step 2-1: calculate Among them, FFT () means fast Fourier transform, LPM () means log-polar mapping, Represents a 7×7 pixel block centered on (i, j) in I ⁽¹⁾ ;

calculate in Represents a 7×7 pixel block centered on (i, j) in I ⁽²⁾ ;

Step 2-2: Let D ⁽¹⁾ (i,j,:)=[F ⁽¹⁾ (1,1),...,F ⁽¹⁾ (1,7),...,F ⁽¹⁾ (7,1 ),...,F ⁽¹⁾ (7,7)];

Let D ⁽²⁾ (i,j,:)=[F ⁽²⁾ (1,1),...,F ⁽²⁾ (1,7),...,F ⁽²⁾ (7,1 ),...,F ⁽²⁾ (7,7)].

After calculation, Possess interest point descriptor D ⁽¹⁾ in ciphertext form, PoI descriptor D ⁽²⁾ in ciphertext form.

(3) Interest point matching:

before calculation, Possess interest point descriptor D ⁽¹⁾ in ciphertext form, Possess interest point descriptor D ⁽²⁾ in ciphertext form, and Both have control parameters α and β, share key set

step 1: Generate a matrix T ⁽¹⁾ of size l _D ×6, where l _D is the number of all (i, j) of D ⁽¹⁾ (i, j,:)≠0, Generate a matrix T ⁽²⁾ of the same size, note that D ⁽¹⁾ (i,j,:)≠0 (i,j) must satisfy D ⁽²⁾ (i,j,:)≠0 at the same time;

Step 2: Use p to represent the index of the currently matched feature point, and Both start matching from the first feature point, that is, let p=1, find the smallest i and corresponding j that satisfy D ⁽¹⁾ (i,j,:)≠0;

Step 3: Find the closest descriptor and its position to the descriptor at (i, j) in ciphertext form, the method is:

Step 3-1: Let T ⁽¹⁾ (p,1)=i,T ⁽¹⁾ (p,2)=j, if q<p exists T ⁽¹⁾ (q,3)=i,T ⁽¹⁾ (q, 4)=j, then let T ⁽¹⁾ (p,3)=T ⁽¹⁾ (q,1),T ⁽¹⁾ (p,4)=T ⁽¹⁾ (q,2),T ^{(1 )} (p,5)=T ⁽¹⁾ (q,5), otherwise let T ⁽¹⁾ (p,5)=inf;

Let T ⁽²⁾ (p,1)=i,T ⁽²⁾ (p,2)=j, if there exists q<p and satisfy T ⁽²⁾ (q,3)=i,T ⁽²⁾ (q, 4)=j, then let T ⁽²⁾ (p,3)=T ⁽²⁾ (q,1),T ⁽²⁾ (p,4)=T ⁽²⁾ (q,2),T ^{(2 )} (p,5)=T ⁽²⁾ (q,5), otherwise let T ⁽¹⁾ (p,5)=0;

Step 3-2: For any (i′,j′), if D ⁽¹⁾ (i′,j′,:)≠0&&(i′>i||j′>j)&&((i′- i) ² +(j′-j) ² )>α, in ciphertext form, judge whether the descriptor at this position is closest to the descriptor at (i,j), the method is:

Step 3-2-1: and interactive computing after calculation have have

Step 3-2-2: and interactive computing after calculation have have

Step 3-2-3: and interactive computing after calculation have have

Step 3-2-4: and interactive computing after calculation have have

Step 3-2-5: calculate

calculate

Step 3-2-6: Calculate u ⁽¹⁾ = (U ⁽¹⁾ (1) + ... + U ⁽¹⁾ (49));

Calculate u ⁽²⁾ = (U ⁽²⁾ (1)+...+U ⁽²⁾ (49));

Step 3-2-7: and Interactive calculation b=SCP(T ⁽¹⁾ (p,5)-u ⁽¹⁾ ,T ⁽²⁾ (p,5)-u ⁽²⁾ ), after calculation and both have b;

Step 3-2-8: If b=1, Let T ⁽¹⁾ (p,3)=, T ⁽¹⁾ (p,4)=j′, T ⁽¹⁾ (p,6)=T ⁽¹⁾ (p,5), T ⁽¹⁾ ( p,5)=u ⁽¹⁾ , Let T ⁽²⁾ (p,3)=i′, T ⁽²⁾ (p,4)=j′, T ⁽²⁾ (p,6)=T ⁽²⁾ (p,5), T ^{(2 )} (p,5)=u ⁽²⁾ .

Step 3-3: and Interactive calculation b=SCP(T ⁽¹⁾ (p,5)×β-T ⁽¹⁾ (p,6),T ⁽²⁾ (p,5)×β-T ⁽²⁾ (p,6)) , after calculating and both have b;

Step 3-4: If b=1, Let T ⁽¹⁾ (p,3)=0,T ⁽¹⁾ (p,4)=0,T ⁽¹⁾ (p,5)=inf,

Let T ⁽²⁾ (p,3)=0, T ⁽²⁾ (p,4)=0, T ⁽²⁾ (p,5)=0.

Step 4: Find the smallest i' and corresponding j' that satisfy D ⁽¹⁾ (i′,j′,:)≠0&&(i′>i||j′>j), if it exists, and All set p=p+1, i=i′, j=j′, return to step 3 and execute again, otherwise go to step 5;

Step 5: In the ciphertext form, reorder the plaintext T corresponding to T ⁽¹⁾ and T ⁽²⁾ according to the size of T(:,5), the method is:

Step 5-1: and All let i=1;

Step 5-2: Find the position of the minimum value of T(i:l _D ,5) in the form of ciphertext, the method is:

Step 5-2-1: and All let θ=i;

Step 5-2-2: For all l _D ≥ i′>i, and Interactive calculation b=SCP(T ⁽¹⁾ (θ,5)-T ⁽¹⁾ (i′,5),T ⁽²⁾ (θ,5)-T ⁽²⁾ (i′,5)), calculate back and Both have b; if b=1, and All let θ=i';

Step 5-3: Exchange T ⁽¹⁾ (θ,:) and T ⁽¹⁾ (i,:); Swap T ⁽²⁾ (θ,:) and T ⁽²⁾ (i,:);

Step 5-4: and All make i=i+1, if i≤l _D , return to step 5-2 and execute again, otherwise go to step 6;

Step 6: For all l _D ≥ i>1, if (T ⁽¹⁾ (i,1)>T ⁽¹⁾ (i,3)||T ⁽¹⁾ (i,2)>T ⁽¹⁾ ( i,4)), then Interaction T ⁽¹⁾ (i,1) and T ⁽¹⁾ (i,3), T ⁽¹⁾ (i,2) and T ⁽¹⁾ (i,4);

If (T ⁽²⁾ (i,1)>T ⁽²⁾ (i,3)||T ⁽²⁾ (i,2)>T ⁽²⁾ (i,4)), then Interaction T ⁽²⁾ (i,1) and T ⁽²⁾ (i,3), T ⁽²⁾ (i,2) and T ⁽²⁾ (i,4);

Step 7: If l _D ≥ i>j satisfies T ⁽¹⁾ (i,:)=T ⁽¹⁾ (j,:), then Delete T ⁽¹⁾ (i,:);

If there exists l _D ≥ i>j satisfying T ⁽²⁾ (i,:)=T ⁽²⁾ (j,:), then Delete T ⁽²⁾ (i,:).

After calculation, There are p matching feature points T ⁽¹⁾ sorted in ciphertext form, Have p sorted matching feature points T ⁽²⁾ in ciphertext form.

(4) Tampering area positioning:

before calculation, p=1, 2, has the sorted matching feature points T ⁽¹⁾ in ciphertext form. Has control parameters γ and δ.

step 1: Generate an all-zero matrix E of size l _m ×l _m , and let i=1;

Step 2: Find the number of points whose distance from T ⁽¹⁾ (i,:) is less than γ, the method is:

Step 2-1: Let n=0, j=i+1, set ε={i};

Step 2-2: Calculate a=(T ⁽¹⁾ (i,1)-T ⁽¹⁾ (j,1)) ² +(T ⁽¹⁾ (i,2)-T ⁽¹⁾ (j,2)) ² ,b =(T ⁽¹⁾ (i,3)-T ⁽¹⁾ (j,3)) ² +(T ⁽¹⁾ (i,4)-T ⁽¹⁾ (j,4)) ² ;

Step 2-3: If a<γ&&b>a/2&&b<2a is satisfied, then Let n=n+1,ε=ε+{j};

Step 2-4: make j=j+1, if j≤l _D , return to step 2-2 and execute again, otherwise go to step 3;

Step 3: If n≥δ, then for all j∈ε, Let E(T ⁽¹⁾ (j,1),T ⁽¹⁾ (j,2))=i, E(T ⁽¹⁾ (j,3),T ⁽¹⁾ (j,4))=- i;

Step 4: Find the smallest circle that includes all points (j,k) where E(j,k)=i, and get all points (j′,k′) inside the circle, Let E(j',k')=i;

Step 5: Find the smallest circle that includes all the points (j,k) of E(j,k)=-i, and get all the points (j′,k′) inside the circle, Let E(j',k')=-i;

Step 6: set i=i+1, if i≤5, return to step 2-3 and execute again, otherwise end the execution;

After calculation, p = 1, 2, has a suspected tampering area E.

3. Client Obtain detection results and areas of suspected tampering

before calculation, Possessing a suspected tampering region E, the concatenated two region distributions are labeled i and -i, i∈{1,...,5}.

step 1: Send E over a secure channel to

After calculation, Possesses suspected tampered area E.

In the above steps, the secure multi-party multiplication protocol (Y ⁽¹⁾ , Y ⁽²⁾ )=SMP(X ₁ ,X ₂ ) is used to calculate The ciphertext matrix X ₁ and The multiplication of the ciphertext matrix X ₂ , and during the calculation will not learn about X ₂ , will not learn about X ₁ , It will not know X ₁ or X ₂ , and none of the three servers will know X ₁ X ₂ , the protocol flow is:

before calculation, With ciphertext matrix X ₁ , With ciphertext matrix X ₂ , and shared key set

step 1: and Select two key matrices K ₁ with the same size as X ₁ (or X ₂ ) according to pre-negotiation,

Step 2: Calculate U ₁ =(X ₁ +K ₂ )/K ₁ , and send U ₁ to

Calculate U ₂ =X ₂ K ₁ , and send U ₂ to

Step 3: Randomly select a random number matrix R of the same size as U ₁ , each element of which is between 0 and 2 ¹⁰ , and calculate V=U ₁ U ₂ +R;

Step 4: Send V to Send R to

Step 5: Let Y ⁽¹⁾ = V;

Let Y ⁽²⁾ = X ₂ K ₂ +R;

After calculation, Having a ciphertext matrix Y ⁽¹⁾ , There is a ciphertext matrix Y ⁽²⁾ , which satisfies Y ⁽¹⁾ −Y ⁽²⁾ =X ₁ X ₂ .

In the above steps, the secure multi-party comparison protocol b=SCP(X ₁ ,X ₂ ) is used for comparison The ciphertext matrix X ₁ and The size of the ciphertext matrix X ₂ , and the calculation process will not learn about X ₂ , will not learn about X ₁ , No X ₁ or X ₂ will be known, the protocol flow is:

before calculation, With ciphertext matrix X ₁ , With ciphertext matrix X ₂ , and shared key set

step 1: and Select two key matrices K ₁ with the same size as X ₁ (or X ₂ ) according to pre-negotiation,

Step 2: Calculate U ₁ =(X ₁ +K ₁ )/K ₂ , and send U ₁ to

Calculate U ₂ =(X ₂ +K ₁ )/K ₂ , and send U ₂ to

Step 3: Calculate U ₁ /U ₂ , if the result is greater than 1, set b=1, otherwise set b=0;

Step 4: send b to and

After calculation, and All have a comparison result b, which satisfies that if b=1, X ₁ >X ₂ , otherwise X ₁ ≤X ₂ .

The image encryption technology selected in this solution can be realized by similar technology, for example, the addition operation is replaced by the subtraction operation to achieve the same effect.

The multiplication protocol and comparison protocol based on multi-party secure computing selected in this solution can be implemented by other types of multi-party security protocols to achieve similar effects, but the calculation efficiency or the number of cloud servers used will change.

The sorting algorithm in this solution can be replaced with any other sorting algorithm to achieve the same effect.

Existing forensics methods for copying image regions can only be performed in plaintext, which will leak image content. In the method of the present invention, the user end can send the image in ciphertext form to the cloud server end, and the cloud server end provides an effective regional copy evidence collection service without knowing the content of the image, and realizes the detection and positioning of the tampering operation. Compared with traditional area copy forensics tools or services, the present invention can provide privacy protection of user image content.

This method is applicable to image digital forensics cloud services, providing safe and reliable regional copy forensics and forensics services for individuals, enterprises, and government departments. This solution is suitable for unreliable cloud service environments.

The above-mentioned embodiment is a preferred embodiment of the present invention, but the embodiment of the present invention is not limited by the above-mentioned embodiment, and any other changes, modifications, substitutions, combinations, Simplifications should be equivalent replacement methods, and all are included in the protection scope of the present invention.

Claims (6)

1. A method for detecting duplication of an image area supporting a privacy protection function, characterized in that it comprises a user terminal 3 independent cloud servers, 2 of which are and For ciphertext storage, 1 server For ciphertext calculation; Use the image encryption algorithm to split the image into two ciphertexts, and send them to and by these two servers and The detection and location of copying and tampering operations in the image area in the ciphertext space are completed through interaction and calculation; finally and Both obtained the suspected tampered area, and send the result to Wherein, the detection method includes the following steps: S1, client Image ciphertext generation steps; including: before calculation, Have an image X to be detected with a size of 1 _m × 1 _m ; Step S1-1: Select a random number matrix R of size 1 _m × 1 _m , each element of which is a random number between 0 and 2 ¹⁰ ; generate 2 copies of image ciphertext: I ⁽¹⁾ = X+R, I ⁽²⁾ = R Step S1-2: Send I ⁽¹⁾ over a secure channel to Send I ⁽²⁾ over a secure channel to After calculation, has ciphertext I ⁽¹⁾ , Possess ciphertext I ⁽²⁾ ; S2, cloud server and The tampering area detection and location steps under the ciphertext space; including: S2-1. Calculating Harris corner points; specifically including: before calculation, With image ciphertext I ⁽¹⁾ , With image ciphertext I ⁽²⁾ , and shared key set Step S2-1-1: calculate calculate Step S2-1-2: and interactive computing after calculation have have Step S2-1-3: and interactive computing after calculation have have Step S2-1-4: and interactive computing after calculation have have Step S2-1-5: and interactive computing after calculation have have Step S2-1-6: calculate where h is the Gaussian filter kernel; calculate Step S2-1-7: and interactive computing after calculation have have Step S2-1-8: and interactive computing after calculation have have Step S2-1-9: and interactive computing after calculation have have Step S2-1-10: and interactive computing after calculation have have Step S2-1-11: and interactive computing after calculation owns A ⁽¹⁾ , has A( ² ); Step S2-1-12: calculate Where t is the Harris corner check constant; calculate Step S2-1-13: and Each generates an all-zero matrix H of size l _m × l _m , which is used to store Harris corner points; for all i and j, Calculate coefficient blocks U ⁽¹⁾ (8i+1: 8i+8, 8j+1: 8j+8) and U ⁽²⁾ (8i+1: 8i+8, 8j+1: 8j+8) in ciphertext form The local minimum value of the corresponding plaintext U(8i+1:8i+8, 8j+1:8j+8), the method is: Step S2-1-13-1: and All let θ _m =8i+1, θ _n =8j+1; Step S2-1-13-2: For all 8i+1<m≤8i+8, 8j+1<n≤8j+8, and Interactive calculation b=SCP(U ⁽¹⁾ (θ _m , θ _n )-U ⁽¹⁾ (m, n), U ⁽²⁾ (θ _m , θ _n )-U ⁽²⁾ (m, n )) , after calculating and Both have b; if b=0, and All let θ _m =m, θ _n =n; Step S2-1-13-3: and Let H(θ _m ,θ _n )=1; After calculation, and Both have the detected Harris corner point H; In the above steps, the secure multi-party multiplication protocol (Y ⁽¹⁾ , Y ⁽²⁾ )=SMP(X ₁ , X ₂ ) is used to calculate The ciphertext matrix X ₁ and The multiplication of the ciphertext matrix X ₂ , and during the calculation will not learn about X ₂ , will not learn about X ₁ , It will not know X ₁ or X ₂ , and none of the three servers will know X ₁ X ₂ , the protocol flow is: before calculation, With ciphertext matrix X ₁ , With ciphertext matrix X ₂ , and shared key set step 1: and Select two key matrices K ₁ with the same size as X ₁ or X ₂ according to pre-negotiation, Step 2: Calculate U ₁ =(X ₁ +K ₂ )/K ₁ , and send U ₁ to Calculate U ₂ =X ₂ K ₁ , and send U ₂ to Step 3: Randomly select a random number matrix R of the same size as U ₁ , each element of which is between 0 and 2 ¹⁰ , and calculate V=U ₁ U ₂ +R; Step 4: Send V to Send R to Step 5: Let Y ⁽¹⁾ = V; Let Y ⁽²⁾ = X ₂ K ₂ +R; After calculation, Having a ciphertext matrix Y ⁽¹⁾ , Have a ciphertext matrix Y ⁽²⁾ , which satisfies Y ⁽¹⁾ -Y ⁽²⁾ = X ₁ X ₂ ; In the above steps, the secure multi-party comparison protocol b=SCP(X ₁ , X ₂ ) is used for comparison The ciphertext matrix X ₁ and The size of the ciphertext matrix X ₂ , and the calculation process will not learn about X ₂ , will not learn about X ₁ , No X ₁ or X ₂ will be known, the protocol flow is: before calculation, With ciphertext matrix X ₁ , With ciphertext matrix X ₂ , and shared key set step 1: and Select two key matrices K ₁ with the same size as X ₁ or X ₂ according to pre-negotiation, Step 2: Calculate U ₁ =(X ₁ +K ₁ )/K ₂ , and send U ₁ to Calculate U ₂ =(X ₂ +K ₁ )/K ₂ , and send U ₂ to Step 3: Calculate U ₁ /U ₂ , if the result is greater than 1, set b=1, otherwise set b=0; Step 4: send b to and After calculation, and Both have a comparison result b, which satisfies that if b=1, X ₁ >X ₂ , otherwise X ₁ ≤X ₂ ; S2-2. Extracting interest point descriptors; S2-3. Interest point matching; S2-4. Tampering with area positioning; S3, client Obtain detection results and areas of suspected tampering. 2. The image region duplication detection method supporting privacy protection function according to claim 1, characterized in that, step S2-2, extracting the POI descriptor specifically comprises: before calculation, With the image ciphertext I ⁽¹⁾ , the detected Harris corner H, With the image ciphertext I ⁽²⁾ , the detected Harris corner H Step S2-2-1: Generate a three-dimensional all-zero matrix D ⁽¹⁾ of size l _m ×l _m ×49, Generate a three-dimensional all-zero matrix D ⁽²⁾ of the same size, both of which are used to store interest point descriptors; Step S2-2-2: For all i and j satisfying H(i, j)=1, calculate the interest point descriptor in ciphertext form, the method is: Step S2-2-2-1: calculate Among them, FFT () means fast Fourier transform, LPM () means log-polar mapping, Represents a 7×7 pixel block centered on (i, j) in I ⁽¹⁾ ; calculate in Represents a 7×7 pixel block centered on (i, j) in I ⁽²⁾ ; Step S2-2-2-2: Let D ⁽¹⁾ (i,j,:)=[F ⁽¹⁾ (1,1),...,F ⁽¹⁾ (1,7),...,F ⁽¹⁾ (7,1 ),...,F ⁽¹⁾ (7,7)]; Let D ⁽²⁾ (i,j,:)=[F ⁽²⁾ (1,1),...,F ⁽²⁾ (1,7),...,F ⁽²⁾ (7,1 ),...,F ⁽²⁾ (7,7)]; After calculation, Possess interest point descriptor D ⁽¹⁾ in ciphertext form, PoI descriptor D ⁽²⁾ in ciphertext form. 3. The image region duplication detection method supporting privacy protection function according to claim 2, characterized in that, step S2-3, the point of interest matching specifically includes: before calculation, Possess interest point descriptor D ⁽¹⁾ in ciphertext form, Possess interest point descriptor D ⁽²⁾ in ciphertext form, and Both have control parameters α and β, share key set Step S2-3-1: Generate a matrix T ⁽¹⁾ of l _D ×6 size, where l _D is the number of all (i, j) of D ⁽¹⁾ (i, j,:)≠0, Generate a matrix T ⁽²⁾ of the same size, note that (i, j) of D ⁽¹⁾ (i, j,:)≠0 must satisfy D ⁽²⁾ (i, j,:)≠0 at the same time; Step S2-3-2: use p to represent the index of the currently matched feature point, and Both start matching from the first feature point, that is, let p=1, find the smallest i and corresponding j that satisfy D ⁽¹⁾ (i, j,:)≠0; Step S2-3-3: Find the closest descriptor and its position to the descriptor at (i, j) in ciphertext form, the method is: Step S2-3-3-1: Let T ⁽¹⁾ (p, 1) = i, T ⁽¹⁾ (p, 2) = j, if there is q<p and satisfy T ⁽¹⁾ (q, 3) = i, T ⁽¹⁾ (q, 4) = j, then let T ⁽¹⁾ (p, 3) = T ⁽¹⁾ (q, 1), T ⁽¹⁾ (p, 4) = T ⁽¹⁾ (q, 2), T ^{(1 )} (p, 5) = T ⁽¹⁾ (q, 5), otherwise let T ⁽¹⁾ (p, 5) = inf; Let T ⁽²⁾ (p, 1) = i, T ⁽²⁾ (p, 2) = j, if there is q < p and satisfy T ⁽²⁾ (q, 3) = i, T ⁽²⁾ (q, 4) = j, then let T ⁽²⁾ (p, 3) = T ⁽²⁾ (q, 1), T ⁽²⁾ (p, 4) = T ⁽²⁾ (q, 2), T ^{(2 )} (p, 5) = T ⁽²⁾ (q, 5), otherwise let T ⁽¹⁾ (p, 5) = 0; Step S2-3-3-2: For any (i′, j′), if D ⁽¹⁾ (i′, j′,:)≠0&&(i′>i||j′>j)&&( (i′-i) ² +(j′-j) ² )>α, in ciphertext form, judge whether the descriptor at this position is closest to the descriptor at (i, j), the method is: Step S2-3-3-2-1: and interactive computing after calculation have have Step S2-3-3-2-2: and interactive computing after calculation have have Step S2-3-3-2-3: and interactive computing after calculation have have Step S2-3-3-2-4: and interactive computing after calculation have have Step S2-3-3-2-5: calculate calculate Step S2-3-3-2-6: Calculate u ⁽¹⁾ = (U ⁽¹⁾ (1) + ... + U ⁽¹⁾ (49)); Calculate u ⁽²⁾ = (U ⁽²⁾ (1)+...+U ⁽²⁾ (49)); Step S2-3-3-2-7: and Interactive calculation b=SCP(T ⁽¹⁾ (p, 5)-u ⁽¹⁾ , T ⁽²⁾ (p, 5)-u ⁽²⁾ ), after calculation and both have b; Step S2-3-3-2-8: If b=1, Let T ⁽¹⁾ (p, 3) =, T ⁽¹⁾ (p, 4) = j', T ⁽¹⁾ (p, 6) = T ⁽¹⁾ (p, 5), T ⁽¹⁾ ( p,5)=u ⁽¹⁾ , Let T ⁽²⁾ (p, 3) = i', T ⁽²⁾ (p, 4) = j', T ⁽²⁾ (p, 6) = T ⁽²⁾ (p, 5), T ^{(2 )} (p,5)=u ⁽²⁾ ; Step S2-3-3-3: and Interactive calculation b = SCP(T ⁽¹⁾ (p, 5) × β-T ⁽¹⁾ (p, 6), T ⁽²⁾ (p, 5) × β-T ⁽²⁾ (p, 6)) , after calculating and both have b; Step S2-3-3-4: If b=1, Let T ⁽¹⁾ (p,3)=0, T ⁽¹⁾ (p,4)=0, T ⁽¹⁾ (p,5)=inf, Let T ⁽²⁾ (p,3)=0, T ⁽²⁾ (p,4)=0, T ⁽²⁾ (p,5)=0; Step S2-3-4: Find the smallest i' and corresponding j' that satisfy D ⁽¹⁾ (i', j',:)≠0&&(i'>i||j'>j), if it exists, and All make p=p+1, i=i', j=j', return to step S2-3-3 to re-execute, otherwise go to step S2-3-5; Step S2-3-5: Reorder the plaintext T corresponding to T ⁽¹⁾ and T ⁽²⁾ according to the size of T (:, 5) in ciphertext form; Step S2-3-6: For all l _D ≥ i > 1, if (T ⁽¹⁾ (i, 1) > T ⁽¹⁾ (i, 3)||T ⁽¹⁾ (i, 2) > T ⁽¹⁾ (i, 4)), then Interaction T ⁽¹⁾ (i,1) and T ⁽¹⁾ (i,3), T ⁽¹⁾ (i,2) and T ⁽¹⁾ (i,4); If (T ⁽²⁾ (i, 1) > T ⁽²⁾ (i, 3)||T ⁽²⁾ (i, 2) > T ⁽²⁾ (i, 4)), then Interaction T ⁽²⁾ (i,1) and T ⁽²⁾ (i,3), T ⁽²⁾ (i,2) and T ⁽²⁾ (i,4); Step S2-3-7: If l _D ≥ i > j satisfies T ⁽¹⁾ (i,:) = T ⁽¹⁾ (j,:), then Delete T ⁽¹⁾ (i,:); If there exists l _D ≥ i > j satisfying T ⁽²⁾ (i,:) = T ⁽²⁾ (j,:), then Delete T ⁽²⁾ (i,:); After calculation, There are p matching feature points T ⁽¹⁾ sorted in ciphertext form, Have p sorted matching feature points T ⁽²⁾ in ciphertext form. 4. the image area duplication detection method that supports privacy protection function according to claim 3, is characterized in that, under the ciphertext form, to T ⁽¹⁾ and T ⁽²⁾ corresponding plaintext T according to T (:, 5) To reorder by size, the method is: Step S2-3-5-1: and All let i=1; Step S2-3-5-2: obtain the position of the minimum value of T (i: l _D , 5) under the ciphertext form, the method is: Step S2-3-5-2-1: and All let θ=i; Step S2-3-5-2-2: For all l _D ≥ i′>i, and Interactive calculation b = SCP(T ⁽¹⁾ (θ, 5) - T ⁽¹⁾ (i′, 5), T ⁽²⁾ (θ, 5) - T ⁽²⁾ (i′, 5)), computing back and Both have b; if b=1, and All let θ=i'; Step S2-3-5-3: Swap T ⁽¹⁾ (θ,:) and T ⁽¹⁾ (i,:); Swap T ⁽²⁾ (θ,:) and T ⁽²⁾ (i,:); Step S2-3-5-4: and Let i=i+1, if i≤l _D , return to step S2-3-5-2 and execute again, otherwise go to step S2-3-6. 5. The image region duplication detection method supporting privacy protection function according to claim 3, characterized in that, step S2-4, tampering region location specifically comprises: before calculation, p=1, 2, has the sorted matching feature points T ⁽¹⁾ in ciphertext form, have control parameters γ and δ; Step S2-4-1: Generate an all-zero matrix E of size l _m ×l _m , and let i=1; Step S2-4-2: Find the number of points whose distance from T ⁽¹⁾ (i,:) is less than γ, the method is: Step S2-4-2-1: Let n=0, j=i+1, set ε={i}; Step S2-4-2-2: Calculate a = (T ⁽¹⁾ (i, 1) - T ⁽¹⁾ (j, 1)) ² + (T ⁽¹⁾ (i, 2) - T ⁽¹⁾ (j, 2)) ² , b = (T ⁽¹⁾ (i, 3) - T ⁽¹⁾ (j, 3)) ² + (T ⁽¹⁾ (i, 4) - T ⁽¹⁾ (j, 4)) ² ; Step S2-4-2-3: If a<γ&&b>a/2&&b<2a is satisfied, then Let n=n+1, ε=ε+{j}; Step S2-4-2-4: set j=j+1, if j≤l _D , return to step S2-4-2-2 and execute again, otherwise go to step S2-4-3; Step S2-4-3: If n≥δ, then for all j∈ε, Let E(T ⁽¹⁾ (j,1),T ⁽¹⁾ (j,2))=i, E(T ⁽¹⁾ (j,3),T ⁽¹⁾ (j,4))=- i; Step S2-4-4: Find the smallest circle that includes all points (j, k) where E(j, k) = i, and get all points (j', k') that are inside the circle, Let E(j', k') = i; Step S2-4-5: Find the smallest circle that includes all points (j, k) where E(j, k) = -i, and get all points (j', k') that are inside the circle, Let E(j', k')=-i; Step S2-4-6: set i=i+1, if i≤5, return to step S2-4-2-3 and execute again, otherwise end the execution; After calculation, p=1, 2, has the suspected tampering area E. 6. The image region duplication detection method supporting privacy protection function according to claim 5, characterized in that, in step S3, the user end Obtaining detection results and suspected tampering areas specifically include: before calculation, Possess a suspected tampering area E, and the spliced two area distributions are marked as i and -i, i∈{1,...,5}; Step S-3-1: Send E over a secure channel to After calculation, Possesses suspected tampered area E.